A reconstruction-based algorithm for classification rules hiding

Data sharing between two organizations is common in many application areas e.g. business planing or marketing. Useful global patterns can be discovered from the integrated dataset. However, some sensitive patterns that should have been kept private could also be discovered. In general, disclosure of sensitive patterns could decrease the competitive ability of the data owner. Therefore, sensitive patterns should be hidden before data sharing starts. To address this problem, released datasets must be modified unavoidably. However, if the overall characteristics of the dataset can be maintained, the dataset is still usable perfectly. Therefore, not only the privacy should be concerned, but also the usability. In this paper, we propose a new algorithm to preserve the privacy of the classification rules by using reconstruction technique for categorical datasets. Firstly, all discovered classification rules in the released dataset are presented to the data owner to identify sensitive rules that should be hidden. Subsequently, remained non-sensitive rules along with extracted characteristics information of the dataset are used to build a decision tree. Finally, the new dataset which contains only non-sensitive classification rules is reconstructed from the tree. From empirical studies, our algorithm can preserve the privacy effectively. Additionally, the usability of the datasets can also be preserved.